You can ask ChatGPT or Claude to tidy merchant names, suggest categories and point out possible duplicates, as long as you remove account numbers, names and phone numbers first and check the result before it goes back into your sheet. Treat the assistant as a fast first pass for text, not as a calculator or an auditor.
What should you remove before pasting anything?
- Bank account and card numbers, including partial ones that identify the account.
- Your name, your family's names and phone numbers inside UPI descriptions.
- UPI IDs, email addresses and home addresses.
- Running balance columns. The assistant does not need them to categorise a row.
- Anything that reveals your employer or a client, if that is confidential.
Do this in a copy. Use Find and Replace to swap a name for the word Person and delete the balance column. Then paste only 30 or so rows to see how it behaves before sending more.
Which prompts work for messy descriptions?
Be specific about the output format. A vague request returns a vague answer.
- Clean names: I will paste bank statement descriptions from India. For each line, return the likely merchant name in plain words, one per line, in the same order. If you cannot tell, write Unknown. Do not guess.
- Suggest categories: Here is my list of categories: Groceries, Eating out, Transport and fuel, Utilities, Health, Shopping, Other. For each row below, give one category from this list only. If none fits, write Other.
- Spot duplicates: In the rows below, list any that have the same amount and a similar merchant within two days. Show the row numbers only. Do not remove anything.
- Explain a sum: Based on the rows below, which five merchants account for the most spending? Show the amounts you used so I can check them.
The phrases about not guessing and keeping to a list matter. Without them, assistants tend to invent plausible-sounding categories.
Can the assistant add up my spending?
It can, but do not rely on it. Language models can slip on arithmetic, especially with long columns of numbers. Use your sheet for totals: =SUMIFS($C$2:$C$500,$B$2:$B$500,"Groceries"). Use the assistant for judgement tasks, such as what a messy description probably is, and always compare its row count with yours. If you pasted 30 rows and got 28 back, find the missing two before trusting anything.
How do you put the results back safely?
- Ask for the output in the same row order as your input.
- Paste it into a new column beside the original, never over it.
- Sort or filter by the new category and skim the groups for odd entries.
- Keep the original column until you are satisfied.
When is a tracker a better fit?
If you clean the same kind of text every month, the cleaning belongs in the tool that holds the data. SpendixAI categorises in stages: your own rules first, then your history and similar merchants, and only if you have switched AI on, an AI model for what is left. In that last step it uses the normalised merchant name and an amount band, not your name, account numbers, full statement text or balances. You can switch AI off per feature, set a monthly AI budget or turn it off entirely, in which case unmatched items are held for review. Features such as reading an unfamiliar statement layout do process the uploaded document.
You can also connect Claude or ChatGPT to your SpendixAI workspace over MCP. You choose a scope, can start read-only and can disconnect at any time. That lets you ask questions of imported data without pasting rows into a chat.
Quick checklist
- Work on a copy.
- Strip personal identifiers.
- Test with 30 rows first.
- Compare row counts.
- Let formulas, not the assistant, do the sums.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


